A Probabilistic Model for Simulating Long-Term Wind-Power Output Scott Kennedy and Peter Rogers Environmental Engineering, Division of Engineering and Applied Sciences, Harvard University, 128 and 116 Pierce Hall, Harvard University, Cambridge, MA 0 2138, USA Email addresses: <skennedy@fas.harvard.edu> and <rogers@deas.harvard.edu> ABSTRACT This paper describes a chronological wind-plant simulation model for use in long-term energy resource planning.The model generates wind-power time series of arbitrary length that accurately reproduce short-term (hourly) to long-term (yearly) statistical behaviour. The modelling objective and methodology differ from forecasting models, which focus on minimizing prediction error. In the present analysis, periodic cycles are isolated from historical wind-speed data from a known local site and combined with a first-order autoregressive process to produce a wind-speed time series model. Corrections for negative wind-speed values and spatial smoothing for geographically disperse wind turbines are discussed. The resulting model is used to simulate the output from a hypothetical offshore wind-plant south of Long Island, New York. Modelled differences of power output between individual turbinesresultfrom wind speedvariability;wake effectsare not consideredin this analysis. 1. INTRODUCTION The increasing contribution of wind-power to the global electricity supply motivates a need for more widely applicable methodologies for evaluating the long-term costs and benefits of wind-power development. Benefits, in particular, are difficult to measure, because they depend on the character of the power system within which wind plantsoperate.Fuel savings, capacity displacement,and avoided emissionscan only be precisely determined by analyzing the stochastic interaction between wind-power output,electricity demand, and conventional generatordispatch.Probabilisticmodelsthatgenerate chronological time seriesof windplant output can be very useful for this purpose. By using chronological data, as opposed to static probability distributions, temporal variability is preserved. The correlation between wind- power availability and time dependent phenomena, such as electricity demand or generator dispatch,can then be examined directly by simulation.In thispaper,a new probabilistic model for simulating wind-power output from a large array of wind turbine generators(W TG’s) or a group of arrays is introduced. Real recorded data for very large windfarms are rare, and so simulated data are useful. In this analysis, modelled differences of power output between individual turbines result from wind speed variability;wake effectsare not considered. W hile there has been considerable effort and a variety of statistical techniques applied to wind-power output models, the majority of such work has focused on short-term forecasting, as opposed to long-term planning. Giebel [2001] reviews the major categories of short-term forecasting models, including persistence models, neural networks, the Risø model, and others. These models are useful for the operational management of a power system because W IND E NGINEERING V OLUME 27, NO . 3, 2003 PP 167–181 167